A Multi Resolution Channel Structure Learning Estimation Method of Geometry Based Stochastic Model W

Vehicle-to-vehicle (V2V) communication is the most typical application of the Internet of Vehicles, and it has many envisaged applications, such as driverless driving, collision warning and addressing traffic congestion. The realization of future intelligent transportation systems (ITSs) will require V2V communication technology, which in turn relies on accurate V2V channel estimation. As the possibilities that 5G technologies can be used for V2V communications are increasingly being explored, corresponding channel models used to describe the characteristics of V2V channels are also being updated, so the new channel estimation algorithms needed to be developed. Although the channel estimation method based on Convolutional Neural Network (CNN) has achieved remarkable success in communication problems in recent years, it cannot be well adapted to V2V channels under different scenarios and different modeling methods. In this article, a method is proposed for learning and estimating the channel structure in the 5G NR downlink that can adapt to the V2V channel under multi-scenes. The inherent block sparsity characteristics of the channel structure is adopted to combine the group lasso Alternating Direction Method of Multipliers (ADMM) algorithm with CNN, and residual dense network is employed to estimate and refine the V2V channel structure. Furthermore, the classifier and interpolation with pooling method are used to estimate the V2V channel structure under multi-scene and multi-resolution. Simulation results show that the performance of this method is better than other deep learning based estimation algorithms.